A novel neural controller for robot manipulator

The success of CMAC (cerebellar model articulation control) for real-time dynamic manipulator control has been exploited by Miller etc. Fundamentally, this kind of neural network deals with discretized state variables and needs relatively large memory to store data. Impressed with the backpropagation´s potential in learning complicated nonlinear mappings, Xu, proposed LBP (localized backpropagation network), which organizes many localized BP subnets into a whole network. In this paper, we discuss its principal further and the practicability of this kind of network in real-time robot manipulator control is also investigated. Simulation results prove this neural network´s architecture has good prospects for applications